StreetFair AI
Introduce a conversational way for homeowners to find the best home service pro for their project.
My Role
Lead product designer and user researcher
Opportunity
Users struggle with knowing which kind of pro would be best for their specific project need. Can AI get a user to a service category and then identify a single pro better than the existing experience?
Explore the problem space
Research showed users struggled to find the right service category, and even when they did, they weren’t sure if pros had the specific skills for their project.
We tested AI by adding colloquial search suggestions. Users liked it, but because it felt too similar to the existing search, they didn’t notice it was AI-powered.
2. Encourage conversational search (over keyword stuffing)
The first mobile concepts were well received after the user learned that their search would initiate an AI-powered chat. We missed the mark communicating that the search was AI-powered (and could free them to be more exploratory and vague in their initial input).
Versions 2–3 tested the AI search input’s home placement. Users picked the best spot and pushed us to make its AI nature obvious, with suggestions visible on first impression.
3. Optimize chat-to-app transitions
Given that the search spawned the chat, determining how the user could interact with and return to an ongoing chat was the next challenge to solve. Many users preferred to browse after the initial matching had been made.
While we, and our users, we’re not the biggest fan of a canvas style approach for desktop, user’s did appreciate and expect to be able to continue interacting with the chat to refine results further.
4. Meet the users expectations for a chat experience
Users expected to be conversed with. Scrolling examples and a welcoming message did the trick.
Users liked getting one primary pro suggestion with a few alternatives. We also added follow‑up questions when details could change the recommendation, for example suggesting a Handyman instead of Siding or Exterior for small siding damage.
Outcome
Project is ongoing as of Aug 2025! So far user reception to the prototypes has been great.
Next steps
While feature is currently in design and being tested with users, engineering is set to explore building a model context protocol that would allow a designated LLM to understand our database schema. We’ll then begin testing interactions and putting guardrails in place to keep user conversations on track.